snudatalab / coheat

CoHeat

This project is a PyTorch implementation of "Cold-start Bundle Recommendation via Popularity-based Coalescence and Curriculum Heating", which is published at The Web Conference 2024.

Prerequisties

The implementation is based on Python 3.10 and PyTorch 2.0.1 A complete list of required packages can be found in the requirements.txt file. Please install the necessary packages before running the code.

Datasets

We use 3 datasets in our work: Youshu, NetEase, and iFashion. The preprocessed dataset is included in the repository: ./data. We separate the dataset into three scenarios: cold, warm, and all.

Configuration

To customize the configuration, please edit the ./src/config.yaml file. For guidance on setting the hyperparameters, please refer to our paper.

Running the code

To execute the code, use the command python main.py with the arguments --data and --seed. For convenience, we provide a demo.sh script that reproduces the experiments presented in our work.

Citation

Please cite this paper when you use our code.

@inproceedings{conf/www/JeonLYK24,
  author    = {Hyunsik Jeon and
               Jong-eun Lee and
               Jeongin Yun and
               U Kang},
  title     = {Cold-start Bundle Recommendation via Popularity-based Coalescence and Curriculum Heating},
  booktitle = {WWW},
  year      = {2024},
}

coheat's Issues

Code for Data Partitioning

How can data be divided into 'warm' and 'cold'? Could you provide the code if possible? Thank you very much.